1 datasets.bundles.deep_scan

datasets.bundles.deep_scan

Opt-in generation-guarded dataset array statistics and target inventory.

1.1 Classes

Name Description
DatasetBundleDeepRequest Complete generation-bound request for the opt-in deep bundle scan.
DatasetBundleDeepStatistics Frozen deep-scan statistics and its already-acquired inventory report.
DatasetCorpusRootDisk Bounded whole-root filesystem observation with its generation identity.
RootDeepAcquisition Detached numeric root evidence, independent of selected rollout membership.
RootDiskCost Whole physical-root cost; unavailable totals carry an explicit reason.

1.1.1 DatasetBundleDeepRequest

datasets.bundles.deep_scan.DatasetBundleDeepRequest(selection, generation)

Complete generation-bound request for the opt-in deep bundle scan.

1.1.1.1 Methods

Name Description
digest Return the deterministic cache and retained-state identity.
1.1.1.1.1 digest
datasets.bundles.deep_scan.DatasetBundleDeepRequest.digest()

Return the deterministic cache and retained-state identity.

1.1.2 DatasetBundleDeepStatistics

datasets.bundles.deep_scan.DatasetBundleDeepStatistics(
    statistics,
    target_inventory,
)

Frozen deep-scan statistics and its already-acquired inventory report.

1.1.2.1 Methods

Name Description
to_jsonable Return a fresh full app/export projection without reacquiring evidence.
1.1.2.1.1 to_jsonable
datasets.bundles.deep_scan.DatasetBundleDeepStatistics.to_jsonable()

Return a fresh full app/export projection without reacquiring evidence.

1.1.3 DatasetCorpusRootDisk

datasets.bundles.deep_scan.DatasetCorpusRootDisk()

Bounded whole-root filesystem observation with its generation identity.

1.1.4 RootDeepAcquisition

datasets.bundles.deep_scan.RootDeepAcquisition(
    generation,
    inventory,
    gt_opportunities,
)

Detached numeric root evidence, independent of selected rollout membership.

The root-only generation binds the acquired rows to the exact selected entry and publication metadata. Bundle reports attach their own request provenance when reusing this evidence; no live reader or mutable array is retained.

1.1.4.1 Attributes

Name Description
generation Root-only selected-entry fingerprint and publication metadata checked before and after acquisition;
inventory Detached complete detected/GT target-population scan in sample-index/row order. Each population
gt_opportunities Frozen per-sample, physical-scene and split counts of finite, non-padding persisted GT-OBB rows.

1.1.5 RootDiskCost

datasets.bundles.deep_scan.RootDiskCost()

Whole physical-root cost; unavailable totals carry an explicit reason.

1.2 Functions

Name Description
acquire_root_deep Acquire complete root evidence once under a root-only generation guard.
compute_dataset_bundle_deep_statistics Compatibility facade returning the historical deep-statistics projection.
inspect_dataset_bundle_deep Acquire one generation-bound deep statistics and target-inventory product.
measure_dataset_corpus_root_disk Observe whole referenced-root file bytes on explicit user dispatch.
scan_root_gt_obb_target_opportunities Count materialized finite, non-padding GT-OBB rows in a VIN root store.

1.2.1 acquire_root_deep

datasets.bundles.deep_scan.acquire_root_deep(request)

Acquire complete root evidence once under a root-only generation guard.

1.2.2 compute_dataset_bundle_deep_statistics

datasets.bundles.deep_scan.compute_dataset_bundle_deep_statistics(selection)

Compatibility facade returning the historical deep-statistics projection.

1.2.3 inspect_dataset_bundle_deep

datasets.bundles.deep_scan.inspect_dataset_bundle_deep(
    request,
    *,
    root_acquisition=None,
)

Acquire one generation-bound deep statistics and target-inventory product.

Supplied detached root evidence must match the request’s exact root entry fingerprint and contain no rollout dependencies. Its numeric rows are reused, while the report receives this bundle’s request provenance. Full bundle currentness is checked before and after either acquisition path.

1.2.4 measure_dataset_corpus_root_disk

datasets.bundles.deep_scan.measure_dataset_corpus_root_disk(
    evidence,
    *,
    max_entries=200000,
    time_limit_seconds=15.0,
)

Observe whole referenced-root file bytes on explicit user dispatch.

Canonical physical roots count once, including all their unselected samples. This measures regular-file lengths, not selected-row cost, allocated blocks or process RAM. One shared entry/time budget bounds traversal across roots. Internal symlinks, special files, read failures and exhausted bounds produce unavailable totals rather than partial values presented as complete costs. Generation guards reject root/shard replacement before and after observation.

1.2.5 scan_root_gt_obb_target_opportunities

datasets.bundles.deep_scan.scan_root_gt_obb_target_opportunities(root_store)

Count materialized finite, non-padding GT-OBB rows in a VIN root store.

This opt-in scan reads only persisted gt.obbs numeric blocks. It never falls back to raw ATEK snippets. Counts represent GT label/evaluation OBB rows available as potential target evidence, not actor-visible proposals, selector-admitted target tasks, or trainable rollout targets.

1.2.5.1 Parameters

Name Type Description Default
root_store Path | str Immutable VIN offline-store directory. required

1.2.5.2 Returns

Name Type Description
dict[str, Any] JSON-compatible availability, total, and per-sample/scene/split counts.
dict[str, Any] An unavailable result includes an exact reason and no inferred count.
Back to top